Comment 2 for bug 1068843

Gavin and I just talked about how to fix this. We looked at various ways of pushing files out from the region controller to the clusters and they all have varying orders of complexity. Given the timescales involved we need a simple solution.

So, we'll make maas-import-pxe-files able to run on the cluster controller such that it pulls kernel/initrd/ephemeral files from a squid proxy running on the region controller. The maas-import-pxe-files script sets up iscsi targets when it runs.

This also means the boot image reporting job has to run on all the clusters, and the database needs to identify which cluster is missing its boot files.

Points to note:
 * check that squid's max object size is big enough to cache the large ISOs
 * Ensure that during PXE boot we point the iSCSI config in kernel params to the cluster's address